1/*
2 * Copyright (C) 2009 SUSE Linux Products GmbH. All rights reserved.
3 *
4 * Authors:
5 *     Alexander Graf <agraf@suse.de>
6 *     Kevin Wolf <mail@kevin-wolf.de>
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License, version 2, as
10 * published by the Free Software Foundation.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
20 */
21
22#include <linux/kvm_host.h>
23
24#include <asm/kvm_ppc.h>
25#include <asm/kvm_book3s.h>
26#include <asm/mmu-hash64.h>
27#include <asm/machdep.h>
28#include <asm/mmu_context.h>
29#include <asm/hw_irq.h>
30#include "trace_pr.h"
31
32#define PTE_SIZE 12
33
34void kvmppc_mmu_invalidate_pte(struct kvm_vcpu *vcpu, struct hpte_cache *pte)
35{
36	ppc_md.hpte_invalidate(pte->slot, pte->host_vpn,
37			       pte->pagesize, pte->pagesize, MMU_SEGSIZE_256M,
38			       false);
39}
40
41/* We keep 512 gvsid->hvsid entries, mapping the guest ones to the array using
42 * a hash, so we don't waste cycles on looping */
43static u16 kvmppc_sid_hash(struct kvm_vcpu *vcpu, u64 gvsid)
44{
45	return (u16)(((gvsid >> (SID_MAP_BITS * 7)) & SID_MAP_MASK) ^
46		     ((gvsid >> (SID_MAP_BITS * 6)) & SID_MAP_MASK) ^
47		     ((gvsid >> (SID_MAP_BITS * 5)) & SID_MAP_MASK) ^
48		     ((gvsid >> (SID_MAP_BITS * 4)) & SID_MAP_MASK) ^
49		     ((gvsid >> (SID_MAP_BITS * 3)) & SID_MAP_MASK) ^
50		     ((gvsid >> (SID_MAP_BITS * 2)) & SID_MAP_MASK) ^
51		     ((gvsid >> (SID_MAP_BITS * 1)) & SID_MAP_MASK) ^
52		     ((gvsid >> (SID_MAP_BITS * 0)) & SID_MAP_MASK));
53}
54
55
56static struct kvmppc_sid_map *find_sid_vsid(struct kvm_vcpu *vcpu, u64 gvsid)
57{
58	struct kvmppc_sid_map *map;
59	u16 sid_map_mask;
60
61	if (kvmppc_get_msr(vcpu) & MSR_PR)
62		gvsid |= VSID_PR;
63
64	sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
65	map = &to_book3s(vcpu)->sid_map[sid_map_mask];
66	if (map->valid && (map->guest_vsid == gvsid)) {
67		trace_kvm_book3s_slb_found(gvsid, map->host_vsid);
68		return map;
69	}
70
71	map = &to_book3s(vcpu)->sid_map[SID_MAP_MASK - sid_map_mask];
72	if (map->valid && (map->guest_vsid == gvsid)) {
73		trace_kvm_book3s_slb_found(gvsid, map->host_vsid);
74		return map;
75	}
76
77	trace_kvm_book3s_slb_fail(sid_map_mask, gvsid);
78	return NULL;
79}
80
81int kvmppc_mmu_map_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *orig_pte,
82			bool iswrite)
83{
84	unsigned long vpn;
85	pfn_t hpaddr;
86	ulong hash, hpteg;
87	u64 vsid;
88	int ret;
89	int rflags = 0x192;
90	int vflags = 0;
91	int attempt = 0;
92	struct kvmppc_sid_map *map;
93	int r = 0;
94	int hpsize = MMU_PAGE_4K;
95	bool writable;
96	unsigned long mmu_seq;
97	struct kvm *kvm = vcpu->kvm;
98	struct hpte_cache *cpte;
99	unsigned long gfn = orig_pte->raddr >> PAGE_SHIFT;
100	unsigned long pfn;
101
102	/* used to check for invalidations in progress */
103	mmu_seq = kvm->mmu_notifier_seq;
104	smp_rmb();
105
106	/* Get host physical address for gpa */
107	pfn = kvmppc_gpa_to_pfn(vcpu, orig_pte->raddr, iswrite, &writable);
108	if (is_error_noslot_pfn(pfn)) {
109		printk(KERN_INFO "Couldn't get guest page for gpa %lx!\n",
110		       orig_pte->raddr);
111		r = -EINVAL;
112		goto out;
113	}
114	hpaddr = pfn << PAGE_SHIFT;
115
116	/* and write the mapping ea -> hpa into the pt */
117	vcpu->arch.mmu.esid_to_vsid(vcpu, orig_pte->eaddr >> SID_SHIFT, &vsid);
118	map = find_sid_vsid(vcpu, vsid);
119	if (!map) {
120		ret = kvmppc_mmu_map_segment(vcpu, orig_pte->eaddr);
121		WARN_ON(ret < 0);
122		map = find_sid_vsid(vcpu, vsid);
123	}
124	if (!map) {
125		printk(KERN_ERR "KVM: Segment map for 0x%llx (0x%lx) failed\n",
126				vsid, orig_pte->eaddr);
127		WARN_ON(true);
128		r = -EINVAL;
129		goto out;
130	}
131
132	vpn = hpt_vpn(orig_pte->eaddr, map->host_vsid, MMU_SEGSIZE_256M);
133
134	kvm_set_pfn_accessed(pfn);
135	if (!orig_pte->may_write || !writable)
136		rflags |= PP_RXRX;
137	else {
138		mark_page_dirty(vcpu->kvm, gfn);
139		kvm_set_pfn_dirty(pfn);
140	}
141
142	if (!orig_pte->may_execute)
143		rflags |= HPTE_R_N;
144	else
145		kvmppc_mmu_flush_icache(pfn);
146
147	/*
148	 * Use 64K pages if possible; otherwise, on 64K page kernels,
149	 * we need to transfer 4 more bits from guest real to host real addr.
150	 */
151	if (vsid & VSID_64K)
152		hpsize = MMU_PAGE_64K;
153	else
154		hpaddr |= orig_pte->raddr & (~0xfffULL & ~PAGE_MASK);
155
156	hash = hpt_hash(vpn, mmu_psize_defs[hpsize].shift, MMU_SEGSIZE_256M);
157
158	cpte = kvmppc_mmu_hpte_cache_next(vcpu);
159
160	spin_lock(&kvm->mmu_lock);
161	if (!cpte || mmu_notifier_retry(kvm, mmu_seq)) {
162		r = -EAGAIN;
163		goto out_unlock;
164	}
165
166map_again:
167	hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
168
169	/* In case we tried normal mapping already, let's nuke old entries */
170	if (attempt > 1)
171		if (ppc_md.hpte_remove(hpteg) < 0) {
172			r = -1;
173			goto out_unlock;
174		}
175
176	ret = ppc_md.hpte_insert(hpteg, vpn, hpaddr, rflags, vflags,
177				 hpsize, hpsize, MMU_SEGSIZE_256M);
178
179	if (ret < 0) {
180		/* If we couldn't map a primary PTE, try a secondary */
181		hash = ~hash;
182		vflags ^= HPTE_V_SECONDARY;
183		attempt++;
184		goto map_again;
185	} else {
186		trace_kvm_book3s_64_mmu_map(rflags, hpteg,
187					    vpn, hpaddr, orig_pte);
188
189		/* The ppc_md code may give us a secondary entry even though we
190		   asked for a primary. Fix up. */
191		if ((ret & _PTEIDX_SECONDARY) && !(vflags & HPTE_V_SECONDARY)) {
192			hash = ~hash;
193			hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
194		}
195
196		cpte->slot = hpteg + (ret & 7);
197		cpte->host_vpn = vpn;
198		cpte->pte = *orig_pte;
199		cpte->pfn = pfn;
200		cpte->pagesize = hpsize;
201
202		kvmppc_mmu_hpte_cache_map(vcpu, cpte);
203		cpte = NULL;
204	}
205
206out_unlock:
207	spin_unlock(&kvm->mmu_lock);
208	kvm_release_pfn_clean(pfn);
209	if (cpte)
210		kvmppc_mmu_hpte_cache_free(cpte);
211
212out:
213	return r;
214}
215
216void kvmppc_mmu_unmap_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte)
217{
218	u64 mask = 0xfffffffffULL;
219	u64 vsid;
220
221	vcpu->arch.mmu.esid_to_vsid(vcpu, pte->eaddr >> SID_SHIFT, &vsid);
222	if (vsid & VSID_64K)
223		mask = 0xffffffff0ULL;
224	kvmppc_mmu_pte_vflush(vcpu, pte->vpage, mask);
225}
226
227static struct kvmppc_sid_map *create_sid_map(struct kvm_vcpu *vcpu, u64 gvsid)
228{
229	struct kvmppc_sid_map *map;
230	struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
231	u16 sid_map_mask;
232	static int backwards_map = 0;
233
234	if (kvmppc_get_msr(vcpu) & MSR_PR)
235		gvsid |= VSID_PR;
236
237	/* We might get collisions that trap in preceding order, so let's
238	   map them differently */
239
240	sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
241	if (backwards_map)
242		sid_map_mask = SID_MAP_MASK - sid_map_mask;
243
244	map = &to_book3s(vcpu)->sid_map[sid_map_mask];
245
246	/* Make sure we're taking the other map next time */
247	backwards_map = !backwards_map;
248
249	/* Uh-oh ... out of mappings. Let's flush! */
250	if (vcpu_book3s->proto_vsid_next == vcpu_book3s->proto_vsid_max) {
251		vcpu_book3s->proto_vsid_next = vcpu_book3s->proto_vsid_first;
252		memset(vcpu_book3s->sid_map, 0,
253		       sizeof(struct kvmppc_sid_map) * SID_MAP_NUM);
254		kvmppc_mmu_pte_flush(vcpu, 0, 0);
255		kvmppc_mmu_flush_segments(vcpu);
256	}
257	map->host_vsid = vsid_scramble(vcpu_book3s->proto_vsid_next++, 256M);
258
259	map->guest_vsid = gvsid;
260	map->valid = true;
261
262	trace_kvm_book3s_slb_map(sid_map_mask, gvsid, map->host_vsid);
263
264	return map;
265}
266
267static int kvmppc_mmu_next_segment(struct kvm_vcpu *vcpu, ulong esid)
268{
269	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
270	int i;
271	int max_slb_size = 64;
272	int found_inval = -1;
273	int r;
274
275	/* Are we overwriting? */
276	for (i = 0; i < svcpu->slb_max; i++) {
277		if (!(svcpu->slb[i].esid & SLB_ESID_V))
278			found_inval = i;
279		else if ((svcpu->slb[i].esid & ESID_MASK) == esid) {
280			r = i;
281			goto out;
282		}
283	}
284
285	/* Found a spare entry that was invalidated before */
286	if (found_inval >= 0) {
287		r = found_inval;
288		goto out;
289	}
290
291	/* No spare invalid entry, so create one */
292
293	if (mmu_slb_size < 64)
294		max_slb_size = mmu_slb_size;
295
296	/* Overflowing -> purge */
297	if ((svcpu->slb_max) == max_slb_size)
298		kvmppc_mmu_flush_segments(vcpu);
299
300	r = svcpu->slb_max;
301	svcpu->slb_max++;
302
303out:
304	svcpu_put(svcpu);
305	return r;
306}
307
308int kvmppc_mmu_map_segment(struct kvm_vcpu *vcpu, ulong eaddr)
309{
310	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
311	u64 esid = eaddr >> SID_SHIFT;
312	u64 slb_esid = (eaddr & ESID_MASK) | SLB_ESID_V;
313	u64 slb_vsid = SLB_VSID_USER;
314	u64 gvsid;
315	int slb_index;
316	struct kvmppc_sid_map *map;
317	int r = 0;
318
319	slb_index = kvmppc_mmu_next_segment(vcpu, eaddr & ESID_MASK);
320
321	if (vcpu->arch.mmu.esid_to_vsid(vcpu, esid, &gvsid)) {
322		/* Invalidate an entry */
323		svcpu->slb[slb_index].esid = 0;
324		r = -ENOENT;
325		goto out;
326	}
327
328	map = find_sid_vsid(vcpu, gvsid);
329	if (!map)
330		map = create_sid_map(vcpu, gvsid);
331
332	map->guest_esid = esid;
333
334	slb_vsid |= (map->host_vsid << 12);
335	slb_vsid &= ~SLB_VSID_KP;
336	slb_esid |= slb_index;
337
338#ifdef CONFIG_PPC_64K_PAGES
339	/* Set host segment base page size to 64K if possible */
340	if (gvsid & VSID_64K)
341		slb_vsid |= mmu_psize_defs[MMU_PAGE_64K].sllp;
342#endif
343
344	svcpu->slb[slb_index].esid = slb_esid;
345	svcpu->slb[slb_index].vsid = slb_vsid;
346
347	trace_kvm_book3s_slbmte(slb_vsid, slb_esid);
348
349out:
350	svcpu_put(svcpu);
351	return r;
352}
353
354void kvmppc_mmu_flush_segment(struct kvm_vcpu *vcpu, ulong ea, ulong seg_size)
355{
356	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
357	ulong seg_mask = -seg_size;
358	int i;
359
360	for (i = 0; i < svcpu->slb_max; i++) {
361		if ((svcpu->slb[i].esid & SLB_ESID_V) &&
362		    (svcpu->slb[i].esid & seg_mask) == ea) {
363			/* Invalidate this entry */
364			svcpu->slb[i].esid = 0;
365		}
366	}
367
368	svcpu_put(svcpu);
369}
370
371void kvmppc_mmu_flush_segments(struct kvm_vcpu *vcpu)
372{
373	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
374	svcpu->slb_max = 0;
375	svcpu->slb[0].esid = 0;
376	svcpu_put(svcpu);
377}
378
379void kvmppc_mmu_destroy_pr(struct kvm_vcpu *vcpu)
380{
381	kvmppc_mmu_hpte_destroy(vcpu);
382	__destroy_context(to_book3s(vcpu)->context_id[0]);
383}
384
385int kvmppc_mmu_init(struct kvm_vcpu *vcpu)
386{
387	struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
388	int err;
389
390	err = __init_new_context();
391	if (err < 0)
392		return -1;
393	vcpu3s->context_id[0] = err;
394
395	vcpu3s->proto_vsid_max = ((u64)(vcpu3s->context_id[0] + 1)
396				  << ESID_BITS) - 1;
397	vcpu3s->proto_vsid_first = (u64)vcpu3s->context_id[0] << ESID_BITS;
398	vcpu3s->proto_vsid_next = vcpu3s->proto_vsid_first;
399
400	kvmppc_mmu_hpte_init(vcpu);
401
402	return 0;
403}
404